model-writtenrun 4632 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Mar 2023 ยท 20 entries
- Robert Haas opened #1 by identifying remaining barriers to safely delegated administration: role-attribute escalation, subscriptions, event triggers, reserved connections, and selective database dumps.
- Nathan Bossart proposed a separate reserve for privileged non-superusers in #4, and Robert agreed in #5 that superuser_reserved_connections should remain an emergency reserve after the new pool is exhausted.
- Nathan submitted the reserved-connections implementation in #6, while Robert independently submitted CREATEROLE attribute restrictions in #8, requiring administrators to possess CREATEDB, REPLICATION, or BYPASSRLS before delegating those attributes.
- Robert's review in #7 led Nathan to count free slots only once and default the new reserve to zero in #9, preserving existing defaults and avoiding inconsistent observations between separate counts.
- Robert identified obsolete replication wording in #10, so Nathan separated its correction into a prerequisite patch in #11; Robert chose development-branch-only treatment in #21 to avoid disturbing translations for a minor fix.
- Nathan's CREATEROLE testing in #12 exposed misleading permission errors, and Robert outlined terse primary errors with explanatory details in #19, establishing the next phase of work.
- Tushar found incorrect reserved-role names and tested connection exhaustion in #15, #16, and #17; Nathan standardized connection terminology in #22 and reordered configuration documentation in #24.
- Robert committed the three unchanged components of #24 as the documentation correction 6c1d5ba486, variable rename fe00fec1f5, and reserved-connections feature 6e2775e4d4, announcing completion in #28.
- Nathan noticed the missing catalog-version bump after landing in #29, and Robert confirmed in #30 that he had accidentally left it in his working tree.
- After Robert clarified the combined CREATEROLE and ADMIN OPTION requirements in #32, Tushar reported successful testing with Neha in #33, and Robert committed #8 unchanged as f1358ca52d.
- Nathan started the error-message series in #35; Tom Lane recommended DETAIL rather than HINT in #36, and Alvaro Herrera proposed reusable parameterized strings and clearer attribute terminology in #37, which Nathan incorporated in #38.
- Robert questioned combining password and attribute failures in #39, and Tom supplied wording preserving the own-password exception in #43, leading to the separate password explanation in #44.
- Peter Eisentraut requested broader cleanup in #45, while Nathan, Robert, Tom, and Alvaro agreed through #47, #48, #49, and #50 that messages generally need not repeat superuser's implicit privileges.
- Alvaro requested a more accurate error code for attempts to demote the bootstrap user in #51, and Nathan incorporated that correction alongside a new companion patch covering COPY, replication, backend signaling, configuration access, and other permission failures in #52.
- After wording refinements in #53, Peter distinguished role attributes from grantable privileges in #54, prompting Nathan's terminology changes in #56.
- Peter then objected to inheritance jargon and second-person phrasing in #57, so Nathan restored simpler privilege wording, adopted impersonal explanations, and combined both patches in #59.
- Peter extracted two unchanged code cleanups from #59: the RenameRole simplification became 3b7cd8c690, and centralizing the replication superuser check became 442f870065, as reported in #60.
- Peter also requested final wording improvements and questioned DropRole's mention of CREATEROLE in #60; Nathan demonstrated the earlier CREATEROLE check in #63 and retained that explanation while submitting the revised remainder in #64.
- Peter committed #64 unchanged as de4d456b40 and announced it in #65, with Nathan acknowledging completion in #66.
- By #66, all attached work had landed, while the subscription, event-trigger, selective-dump, and finer-grained role-management ideas raised in #1 remained outside the implemented work in this thread.